Screen-Aware Dictation for Mac: Reply to What You See by Voice

Screen-aware dictation that reads your Mac, not just your voice

Context mode is the only Verba mode that looks before it writes. When you trigger it, Verba captures your current screen, sends that image to a vision-capable model on the AI engine you chose, and uses what it sees as the source material for your spoken instruction. That means you can stop describing the email, the error, or the table on screen and just point at it with your voice. The result is pasted wherever your cursor sits, in any app.

• Captures the screen, then writes from it with a vision model
• Trigger with Fn+X, like every other Verba mode
• Say the instruction, not the context: "reply to this", "explain this error"
• Pastes the finished text into whatever app holds the cursor
• Runs on your own AI: Claude subscription, Anthropic key, OpenRouter, or local Ollama vision model

Reply to what is on screen by voice

The everyday win is replying to things you can see but would rather not retype. Open an email and say "reply, politely decline, suggest next week instead" and Context mode reads the message, drafts the reply in your voice, and drops it in the compose box. Point it at a Slack thread, a support ticket, a code review comment, or a form, and it writes from the real content on screen instead of a vague mental summary you have to dictate first.

• Email replies that already know what the sender wrote
• Answers to a Slack or chat thread you have open
• Filling a form or field based on the data shown around it
• Explaining or rewriting an error, log, or stack trace on screen
• Summarizing a long article or document you are looking at

What Context mode needs to see your screen

Because it reads pixels, Context mode has two requirements the verbatim modes do not. First, macOS Screen Recording permission, which Verba requests so it can capture the screenshot it sends to the model. Second, a vision-capable model on your AI engine, since a text-only model cannot read an image. Both Claude (via your subscription or Anthropic key) and many OpenRouter and Ollama models support vision; pick one of those for Context. Once granted, the screenshot goes only to the engine you configured for rewriting.

• Requires macOS Screen Recording permission to capture the screen
• Requires a vision-capable model on your chosen AI engine
• Works with Claude, Anthropic, OpenRouter, or a local vision model
• Bring-your-own-AI: Verba never makes a billed call on your behalf

How Context mode differs from Verba's other modes

Flow and Polish only transcribe and clean what you say. Intent writes to an instruction but knows nothing about your screen. Translate converts language, and JARVIS (Fn+X's cousin, Action mode) takes real actions across connected apps after you confirm. Context sits between dictation and agent: it does not act on your behalf or touch other apps, it simply reads the screen and writes from it. If you want screen-aware writing without an agent doing anything, Context is the mode. Switch to it with Fn+1..9 like any other, and layer a style with Fn+]/[ if you want a particular tone.